Test and Diagnosis of of ICs Fault coverage (%) 95 9 85 8 75 7 65 97.92 SSL 4,246 Shawn Blanton Professor Department of ECE Center for Silicon System Implementation CMU Laboratory for Integrated Systems Test http://www.ece.cmu.edu/~blanton 99.46 99.93 99.54 99.49 74,5 74,5 74,5 Bridging 94.78 28,564 Gate delay Defect modeling Test methodology development CAD tool development Defect diagnosis 79. Test grading with FAult Tuple SIMulator 74,5 44,798 TSO Courses: 8-24: Fundamentals of Computer Engineering 8-34: Digital Computation 8-44: Verification of Computer Hardware Systems 8-765: Digital Systems Testing and Testable Design µmechanical structure MEMS Test µmechanical structure affected by a defect Defect modeling Built-in self test (BIST) design Test methodology development
Digital Circuit Design Emerging Trends in Electrical and Computer Engineering Shawn Blanton Fall 24 blanton@ece.cmu.edu
Components of Digital Design Function Correctness Digital design Cost Power Speed
Function When we design a circuit to do a particular function, we assume we have primitive gates like AND, OR, etc. So we build bigger things from smaller things. And we have systematic methods (i.e., CAD tools) for building bigger things. y y2 y2 A B C D x = A, B, C, D, x = B, C, D, A, K-maps Quine-McCluskey Boolean Algebra Synplicity etc.
Function cont But what if you wanted to build a fast, 64-bit adder? What about using a ripple-carry adder? Systematic techniques cannot handle some things we want to build. How many rows are in the truth table for a 64-bit adder?..................... : : : : : : : : : :...
Function cont There are many known adder architectures. To learn how to design an adder for your given application, you must see these adders and understand how they work. 32-bit carry-skip adder 4 FAs 4 FAs 4 FAs 4 FAs 4 FAs 4 FAs 4 FAs 4 FAs The same is true for other data-processing circuits: multipliers, dividers, filters, etc.
Customized Floating Point Unit 64-bit floating point 3-bit custom floating Courtesy of Prof. Rob Rutebnar
Cost $$$ Many ideas you may have are likely possible. But as engineers, it must be possible at a reasonable cost. One type of cost has to do with design size. Ideally, you want a circuit implementation to take very little area for a number of reasons: Small circuits means more can fit onto a wafer. Small circuits have a higher yield. wafer Yield = good chips Total chips
Cost $$$ cont Log scale The number of gates available in the same area is doubling every 8 months ( Moore s Law ). Designer productivity is not increasing at the same rate. This leads to the design gap. Solution: Better CAD tools for design and analysis!
Speed Your new adder design may add, but does it add fast enough? Digital design requires you to understand how to accurately predict the speed of your circuit. Lots of factors affect speed: Design structure Fabrication technology Environment Actually, you cannot really predict speed that accurately. 4GHz 3.66GHz 3.GHz 2.6GHz
Power Power is now the number one challenge for digital circuit designers! Why? Many electronic systems are portable. Low power means longer battery life. High power generates lots of heat so designs must be cooled.
Correctness Designers make mistakes. How do you find errors in a design? In 8-24, your method for finding mistakes is either ad hoc or impractical. You need systematic approaches that utilize CAD tools that include simulators, equivalence checkers, model checkers, etc. Big companies still make mistakes. Intel made an error designing a divider circuit. It cost Intel $47M.
Correctness cont Manufacturing is not perfect. How do you find imperfections in the fabricated design? You must test each and every chip to ensure it works.
Test and Diagnosis Test is a binary endeavor Chip doesn t work trash it! Chip works ship it! Diagnosis Why doesn t the chip work??? Required for silicon debug, process tuning, etc.
What do you test for? Test the chip function? Portions of the chip are tested this way. Too costly however for the entire chip. It is more cost-effective to test for the things that can go wrong.
What can go wrong? One or more defects can occur! A defect can be extra or missing material or changes in material properties Defects can affect wires and vias Defects can affect transistors Defect-based test is impractical Too many possible locations (billions) Too many different types (N billions) To keep test manageable, abstraction is necessary
Defect Abstraction Abstraction involves Moving from physical level to higher level Reducing number of defects Simplifying defect behavior Defect abstraction fault model
SSL Fault Model The most widely-use abstraction is single stuck-line (SSL) fault model. First published use was for vacuum tube test in 957 by R. Eldred. SSL fault model properties: a gate-level model assumes gates are unaffected only a single logic line can be faulty faulty line is permanently stuck-at or
Stuck-at Testing The circuit above has 9 logical lines. Each line is susceptible to either a stuck-at- or stuck-at- fault, for a total of 8 faults. Must find tests that distinguish the 8 faulty circuits from the good circuit.
Stuck-at Testing cont d Test set. 2. 3. 4. 2 3 9 9/ 7/ 2/ 9/ 7/ 5/ 8/ / 2/ 3/ 6/
Digital Design Courses Learn more by taking various courses in digital design. 8-24 Fundamentals Advanced digital design 8-34 8-36 Computer-aided design tools and correctness Computer architecture Correctness 8-447 8-44 8-76 Advanced computer-aided design tools 8-765 Testing